5

What is the minimum number of tubes per bit required to store a CPU register in a vacuum-tube computer?

In particular, have any real vacuum-tube computers ever successfully stored more than 1 bit of a CPU register per tube?

Related, but not quite the same questions:

  • How many 6SN7 tubes did it take to store a bit? asks specifically about the 6SN7 tubes

    • answers seem to imply 2 tubes per bit.

  • Answers to Can I build a working(ish) vacuum tube byte? mention

    • 12AX7 tubes (a dual triode); that answer seems to imply 1 tube per bit, and
    • dekatron tubes which can store 1 decimal digit per tube (and so could theoretically store 3 bits per tube); but it's unclear if this is actually usable as a CPU register storage.

  • The Wikipedia Williams Tube article seems to imply it could store 16*16 = 256 bits, but also seems to imply it was not usable as CPU register storage.

David Cary
  • 216
  • 2
  • 7
  • 4
    Not really sure what the question is, as the referenced questions do already answer it. 2 Tiodes or 1 Dual Triode (In the end it does not make any sense to count 'tubes' here as it's about triodes, not how many are packed into a single glass container) or something around 1/3rd using a decatron or 1/256th using a Williams tube. What other answer is there to be expected? (And BTW, as mentioned in the answer, the 6SN7 contains two Triodes, so it's as well only one tube) – Raffzahn Aug 06 '21 at 21:02
  • Storage doesn't know what it is storing, per se. How would a few vacuum tubes know they should not work as "CPU" storage, but only some other kind? – Jon Custer Aug 06 '21 at 21:28
  • 1
    The early history of the physical design of vacuum tubes involved various industry cartels and patent licensing agreements. One consequence was that the number of connection pins per tube was effectively limited to 8 or 9, which meant a limit of two independent triodes per tube. A few specialized devices were manufactured for particular applications which broke that limit (e.g. 12-pin tubes for TV circuit applications), but the inertia of international standardization and interchangeable parts was too strong to be broken. – alephzero Aug 06 '21 at 22:20
  • @JonCuster: I agree that bits are bits, and yet "reading from or writing to the internal registers, normally implemented as flip-flops" in the Wikipedia "Williams Tube" article suggests that there's something I'm missing, hence my question. – David Cary Aug 06 '21 at 22:22
  • @Raffzahn: yes, that would be the answer if I was asking "how many bits per tube if I'm forced to use these particular tubes", but people building such machines are not limited to those particular tubes. Would it help to edit my question to say "Is there some other kind of tube that could store more bits per tube and still be used as a CPU register?" – David Cary Aug 06 '21 at 22:26
  • 1
    @DavidCary Given the relatively low reliability of tubes, there was not much incentive to design something using a single source of non-standard devices, compared with what modern electronics engineers would call "jelly bean" part numbers, obtainable from anywhere in any quantity required. – alephzero Aug 06 '21 at 22:31
  • 2
    @DavidCary Well, there is no logical reason not to make a "hexatron" tube like a dekatron but with 16 electrodes, or whatever - but there were good commercial reasons why nobody actually did it. (And the Williams tube was superseded by magnetic core memory, not by better thermionic devices) – alephzero Aug 06 '21 at 22:39
  • 2
    @DavidCary It feels somehow not fully thought thru. For one, a register is something else than storing a bit. A register contains read and write logic as well as selection and so on. Next, a tube (as in glas tube) is only the container for elements needed, the active elements are triodes. Further how many triodes are packed into a glass tube is not so much a technological issue (As Alephzero already explained very well). And last, even for just a bit storage (aka a Flip Flop) one needs more components than just two triodes. Bottom line, as it stands the question does not make a lot sense. – Raffzahn Aug 06 '21 at 22:40
  • @DavidCary It feels as if that question can be reduced to three questions: a) How many triodes can be put into a tube: Answer: almost unlimited. b) Maximum number put into one: Four (IIRC). c) Number usually put into one: Two. The only way I see to improve the question would be to describe why/for what purpose it is asked. Oh, and speaking of a hexatron, this opens another important issue: technology used. It would not just store bits. A decatron like hexatron works only well in serial architectures - but one quite different from the kind of binary serial we know. – Raffzahn Aug 06 '21 at 22:49
  • 2
    re: dekatron, but it's unclear if this is actually usable as a CPU register storage -- see Harwell WITCH, which used it for primary storage. You would, I suppose, choose decimal storage iff you were building a decimal machine, and not otherwise. – dave Aug 07 '21 at 00:25
  • 2
    You probably need to define "register" as well. In modern use, 'registers" are distinct from "memory". In earlier architectures, "registers" were anything that stored bits, including memory locations identified by "addresses". – dave Aug 07 '21 at 00:32
  • @alephzero: Do you know if any machines were designed to use dynamic (continuously pumped) vacuum rather than static vacuum? I would think that a dynamic vacuum design would make it possible to have a fairly complex but serviceable assembly within a vacuum chamber, if one could power down a system, let in air, open the chamber, perform service, pump down the system, and power it back up. Not sure how vacuum pump technology in the 1940s compared to today, though. – supercat Aug 09 '21 at 14:35
  • Probably not helpful: But I remember reading about using a long-phosphor CRT as a storage mechanism. Data was displayed on the CRT, which due to it's long duration phosphor would 'glow' for some interval, at which time a camera (pointed at the CRT) would 'read' back the data, possibly leading to another refresh cycle of the CRT. This would infer, a single tube (or two if you count the camera) can 'store' data. - but I'm probably not really helping at all. :-) – Geo... Aug 09 '21 at 17:18

1 Answers1

8

The Manchester Small-Scale Experimental Machine used two Williams–Kilburn tubes to hold its registers: one stored a 32-bit accumulator, while another stored the current 32-bit instruction and the current instruction address, or program counter. This was aside from the 32×32 bit Williams–Kilburn tubes used as main memory and display. The Manchester Mark 1 and the derived Ferranti Mark 1 also used the same technology for memory.

The largest proposed thermionic valve memory storage unit was the RCA Selectron at 4096 bits. RCA were never able to produce this reliably in quantity, and by the time the simplified 256 bit Selectron was ready for market, magnetic core had already taken most of the market.

Omar and Lorraine
  • 38,883
  • 14
  • 134
  • 274
scruss
  • 21,585
  • 1
  • 45
  • 113